Bypassing conscious control: Media violence, unconscious imitation, and freedom of speech1
Forthcoming in Does Consciousness Cause Behavior? An investigation of the Nature of Volition, eds. S. Pockett, W. Banks, and S. Gallagher, MIT Press.
'We are what we pretend to be, so we must be careful about what we pretend to be.'
Kurt Vonnegut, Mother Night2
Why does it matter whether and how individuals consciously control their behavior? It matters for many reasons. Here I focus on concerns about social influences of which agents are typically unaware on aggressive behavior.
First, I survey research about the influence of viewing media violence on aggressive behavior. The consensus among researchers is that there is indeed a robust causal influence here.
Second, I explain why, in the context of work in cognitive science and neuroscience on imitation, this influence is not surprising. Indeed, it would have been surprising if aggressive behavior had been immune from general imitative influences. Imitation is a topic of intense contemporary scientific interest and of great importance for understanding what is distinctive about human minds. Recent advances in understanding imitation and related processes shed light on the mechanisms and functions that may underlie the influence of media violence on aggressive behavior. Human beings have a tendency to imitate and assimilate observed or represented behavior, which operates at various levels and is often automatic and unconscious. Automatic imitative influences of which individuals are unaware can compromise their autonomy.
Third, I consider how this bears on the liberal principle of freedom of speech. This principle goes beyond the general liberal principle of freedom of action so long as action does not harm others; speech is given additional, special protection from interference, even when it does harm others. Why? Answers often invoke autonomy. In particular, speech is often assumed to engage autonomous deliberative processes in hearers by which they consciously control their responses to speech. Social influences that bypass autonomy are implicitly assumed to be negligible. But what if they aren’t? Empirical work on imitation is relevant to both the likely effects of speech and to the character of the processes by which its effects are brought about.
As a result, I suggest, liberals should begin to think about responsibility in ecological terms: to rethink issues about the way individual responsibility depends on social environments, and about shared individual and social accountability for the effects of speech when individuals are subject to social influences that compromise their full autonomy. This requires understanding imitation and its implications for issues about media violence, rather than persisting in unscrutinized assumptions about the autonomous character of responses to speech.
Issues about media violence have been relatively neglected in discussions of freedom of speech, compared, say, to pornography and hate speech. This is unfortunate, because media violence raises basic issues about harm to third parties in a simpler way, without also raising the complex issues about paternalism, equality, and silencing effects that are raised by pornography and hate speech. The basic issues are still difficult, and getting clearer about them may help to make progress on all fronts. In particular, there is a large overlap between media violence and pornography, so arguments about media violence may apply to some pornography, in ways that do not depend on the more complex and controversial arguments about equality and silencing effects. My lack of attention here to pornography per se should thus not be taken to imply that I regard my arguments as irrelevant to it. On the contrary.
I want to emphasize that the argument I develop about media violence is not paternalistic: it turns on harm to others, not on harm to self. Opposition to paternalism is irrelevant to it. Moreover, I am not arguing for any particular policy response; that is a difficult further question. Rather, I am trying to open up a set of issues and to excavate connections that need considerable further thought, at the levels of both theory and legal policy.
2. Does exposure to media violence tend to increase aggressive behavior?
2.1. The effects of media violence: consensus and denial. Several hundred research studies have examined the relationship between exposure to violence in the media and aggressive behavior.3 They have measured the effects of violent entertainment and, to a lesser degree, of news reports of violence, and have focussed on whether exposure to media violence tends to increase aggressive behavior. In general, aggressive behavior is defined conservatively in terms of intentional production of physical harm or threats of physical harm.4 Responses have been highly politicised, and proponents of each side have often been regarded as selective in their appeals to evidence.5 Unfortunately, this work has been assessed without putting it into the essential context of our growing knowledge of imitation in general, as I shall do here.
Since the Williams (1981) Report, many liberals have felt able to dismiss as unsubstantiated the thesis that exposure to media violence tends to increases aggression.6 This dismissal is now seriously dated. Surprisingly, there is a strong disconnect between even educated public opinion on this subject and the increasingly convergent majority opinion of researchers.7 This strong and growing consensus has often not been accurately reported in the media, and has not got across to the public. Here are some quotations from recent textbooks:
“…the consensus of research opinion accepts that media violence does cause violence in society” (Howitt 1998, 61, who says that about 80% of the studies support this thesis.)8
“The evidence in support of the hypothesis that viewing violence leads to an increase in aggressive behavior is very strong” (Gunter and McAleer 1997, 114).
“For most investigators of media violence effects…there’s no longer a question about whether viewing violent events in the mass media enhances the likelihood of further aggression.” (Berkowitz 1993, 208)
This consensus view was endorsed in 2000 by the American Psychological Association, the American Academy of Pediatrics, the American Academy of Child and Adolescent Psychiatry, the American Medical Association, the American Academy of Family Physicians, and the American Psychiatric Association (see Bushman & Anderson 2001).
If readers wish to sample a few of the many studies, some suggestions: An influential meta-analysis published in 1994 by Paik and Comstock found a positive significant correlation between exposure to media violence and aggressive behavior, regardless of research method (see also Comstock & Scharrer, 2003; Anderson & Bushman 2002, 2377; Anderson et al 2003). Science recently published a longitudinal study following over 700 families for 17 years (Johnson et al 2002), linking television exposure to subsequent aggression. This study is striking because it shows, contrary to what may be assumed, that the link holds for adolescents and young adults as well as children.9 Alternate explanations that were statistically controlled for and ruled out include childhood neglect, family income, neighborhood violence, parental education, psychiatric disorders, etc.10 Recent evidence shows that exposure to violent video games also tends to increase aggression (Anderson and Bushman 2001).
2.2. Effect sizes. What is the magnitude of the effects of media violence? The effect sizes shown in the 1994 meta-analysis are larger than the effects, for example, of calcium intake on bone mass, or of lead exposure on IQ in children, or of asbestos exposure to cancer—all risks we take seriously (Bushman and Anderson 2001, 480-481; Comstock and Scharrer 2003). Moreover, even small effects are magnified in human significance by the huge viewing audiences of the mass media (as the advertising industry well appreciates). Suppose, for example, that only 10% of viewers who see a given program are influenced to be more aggressive11; this may seem a small effect size, but 10% of 10 million viewers is still a million people. The effects of increased aggression in a million people, some of which threatens or physically harms third parties, may be highly significant in terms of human suffering (Rosenthal 1986). Whether this is a price worth paying is a further question; perhaps it is. But we shouldn’t duck the question.
2.2. Methodologies. The research meta-analysis pulls together is of three main methodological types (Anderson and Bushman 2001, 354). First, experimental studies randomly assign subjects to violent or non-violent media and assess them later for aggression. Lab experiments are not merely correlational, but permit rigorous controls and support causal inferences, according to standard scientific criteria. However, they are criticized as artificial in relation to real-world violent behavior, and for focusing on short-term effects (Paik and Comstock 1994, 517; Geen 1990, 95). The Williams Report pointed out that criminal and antisocial behavior cannot ethically be experimentally produced, so surrogates or related behaviors are often measured; in some cases this involves fictional or pretend contexts, such as hitting a doll, begging the question of the relationship between fictional contexts and real-world aggression (1981, 65-66, 68).12 Field (as opposed to lab) experiments achieve similar controls while making more naturalistic measurements (though the degree of control may not be as high). For example, institutionalised delinquent boys were first assessed for aggressiveness, and then shown violent or non-violent films for several nights. Those shown violent films were subsequently more violent in daily life, not just in the lab. The effect was especially strong for the boys who had previously rated low in aggressiveness (Berkowitz 1993, 207-208).
Second, there are correlational field studies, which assess types of media consumed and their correlation with aggression at a given time. As is well recognized in this literature, correlational studies alone do not establish causation.
Third, longitudinal studies assess correlations between types of media consumed and aggression across time periods. The cross-lagged longitudinal technique used in some longitudinal studies does support causal inferences.
2.3. Correlation vs. causation. It’s worth pausing to focus on the familiar objection that correlation does not show causation, which is often voiced in relation to the thesis that media violence has harmful effects (Renfrew 1997, 161; Williams 1981, 65, 71, 84; see also and cf. Chartrand and Bargh 1999, 895). Perhaps aggressive tendencies cause people to seek out media violence—the ‘reverse hypothesis’--, or perhaps both have some further common cause. It is important to recognize that media violence research addresses this point effectively.
First, triangulation between different research methods and the mutually supporting results they yield in meta-analyses supports the thesis that exposure to media violence is not merely correlated with increased aggression but tends to cause it. Experimental studies with careful controls and manipulations support causal claims, and they converge with the results of more naturalistic fieldwork and correlational studies (Renfrew 1997, 157; Potter 2003, 28-29).13
Second, cross-lagged longitudinal studies also support causal inferences (Eron et al 1972; Renfrew 1997 156, 161). Here, measures of two variables are made at both an earlier and a later time. If the correlation between A1 and B2 is substantially greater than that between B1 and A2, this indicates that A causes B, not vice versa. For example, viewing media violence by 8-year-old boys is strongly correlated with aggressive behavior by those same boys 10 years later, but aggressive behavior at the earlier age is not strongly correlated with viewing media violence 10 years later.14 The pattern holds across all levels of aggressiveness in the earlier age group, so is not due to different levels of aggression in the younger children. The result controlled for other variables such as parental socio-economic status, boys’ IQ, parental aggressiveness, etc. A similar pattern is found again 12 years further on, at age 30, bridging 22 years. The 8-year-old boys with the strongest preference for media violence were most likely to have been convicted of a serious crime by age 30.15 Similar results have been found across five different nations (Berkowitz 1993, 228-229).
As time passes, the relationship between exposure to media violence and aggression may well be reciprocal or circular: exposure may increase aggressive tendencies, which in turn lead those affected to seek out further exposure to media violence. The supposition that those with aggressive tendencies may tend to seek out media violence is consistent with the thesis that media violence increases aggression; aggression and media violence may be in this way mutually reinforcing, in a truly vicious cycle. 16
2.4. Short- vs. long-term effects. It’s helpful to distinguish short-term and long-term influences of media violence. Berkowitz explains the short-term effects in terms of priming (1993, 210ff; see also the discussion below of imitation and the chameleon effect). Several studies show that mere exposure to words connected with aggression primes punitive behavior to a fellow experimental subject. Visual images are especially effective primes. Subjects are more aggressive to someone who has earlier provoked them after they watch a video scene in which a villain gets a deserved beating. Berkowitz also gives the example of suicide contagion: combined evidence from thirty-five US cases shows a significant upward trend in the number of suicides in the month or so after a celebrity’s suicide is reported. There are similar findings for the UK (Berkowitz 1993, 204-205).
Huesmann, an author of the influential 22 year cross-lagged longitudinal study cited above, gives a unified account of short- and long- term cumulative effects in terms of the acquisition of cognitive scripts that serve as guides for behavior over the longer term. Scripts can be acquired through social learning or imitation as well as through direct experience of results, and incorporate procedural know-how as well as declarative knowledge (Huesmann 1998, 89, 97; Huesmann 2005). Predisposing personal factors and environmental context interact through observational and enactive learning to lead to the emergence of aggressive scripts (1998, 96). Repeated exposure to certain behavior patterns does not merely prime short-term copying behavior, but also establishes particular scripts and makes them more readily accessible for use over the longer term. Certain types of media violence are more likely to generate aggressive scripts than others.17 However (as one would expect from the literature surveyed below on the ‘chameleon’ and related effects), exposure to violent media can increase the accessibility of violent behavioral responses in ways that bypass norms, values and attitudes over which the agent has conscious control (Comstock and Scharrer 2003); such influences can operate even though the agent ‘officially’ has a negative attitude toward violent behavior (see also Wilson & Brekke, 1994). Both short-term priming and longer-term script effects can be triggered, without awareness on the part of the subject, merely by the presence of apparently irrelevant contextual cues: children exposed to a violent scene in which a walkie-talkie features have been shown to be more likely to behave aggressively later in the presence of a walkie-talkie; the color of a room in which violence is observed can have a similar effect (Huesmann 1998, 82, 83, 97, 98; Berkowitz 1993, 222). While priming effects and script activation often can be evaluated and inhibited, it is a further question how the evaluative and inhibitory scripts are acquired and activated. Scripts are more likely to be controlled while they are first being established, but in the longer term and with maturity, may become automatic (Huesmann 1998, 90). As well as activating aggressive scripts, viewing violence may disinhibit violence through desensitization and changing what is regarded as normal. Children who have just viewed a violent movie are slower than controls to intervene in serious fighting between other children (Berkowitz 1993, 223-224). One often hears the view that media violence may have a cathartic effect, in defusing pent up aggression. Unfortunately, the evidence simply does not support a catharsis effect (Huesmann and Taylor, 2003; Potter 2003, 39).
In summary, the evidence shows that exposure to media violence causes an increased tendency to aggressive behavior of significant effect size across the population of viewers, in both the short and long terms.
By what mechanisms does exposure to media violence have these effects? I suggest that these may include the evolutionarily useful tendency of humans to imitate actions they observe others performing, which operates in part automatically and unconsciously.18 Moreover, to the extent an automatic, unconscious tendency19 to imitate explains the influence of exposure to media violence, we cannot expect introspective assessments of this influence by those subject to it to be accurate (see Potter, 2003, ch. 2, on the myth that “it may affect others but it doesn’t affect me”). Let’s next consider what has been learned about human imitative tendencies.